A variety of challenges exist in the design of systems that can be used to host a wide variety of computer-generated actors (CGAs) that possess believable behaviors. The challenges arise in the areas of system architecture and design, knowledge-base design, decision-making mechanisms, and the distributed virtual environment (DVE) network interface. These challenges are especially significant if the DVE is to be used for training, because accurate training is essential to the ready application of training experience to real-world situations. The project described in this paper was undertaken to improve the quality of threat CGAs in DVEs utilized for aircrew training. In this paper, we describe the system and the reasons for its genesis. We present the system requirements, system architecture, component-wise decomposition of the system design, and structure of the major components of the decision mechanism. We conclude with a summary of our results to date and recommendations for further research.